Computational Methods and Production Engineering: Research and Development
()
About this ebook
Computational Methods and Production Engineering: Research and Development is an original book publishing refereed, high quality articles with a special emphasis on research and development in production engineering and production organization for modern industry. Innovation and the relationship between computational methods and production engineering are presented.
Contents include: Finite Element method (FEM) modeling/simulation; Artificial neural networks (ANNs); Genetic algorithms; Evolutionary computation; Fuzzy logic; neuro-fuzzy systems; Particle swarm optimization (PSO); Tabu search and simulation annealing; and optimization techniques for complex systems.
As computational methods currently have several applications, including modeling manufacturing processes, monitoring and control, parameters optimization and computer-aided process planning, this book is an ideal resource for practitioners.
- Presents cutting-edge computational methods for production engineering
- Explores the relationship between applied computational methods and production engineering
- Presents new innovations in the field
- Edited by a key researcher in the field
Related to Computational Methods and Production Engineering
Related ebooks
Manufacturing Engineering Education Rating: 0 out of 5 stars0 ratingsStandards, Quality Control, and Measurement Sciences in 3D Printing and Additive Manufacturing Rating: 0 out of 5 stars0 ratingsLife Cycle Sustainability Assessment for Decision-Making: Methodologies and Case Studies Rating: 0 out of 5 stars0 ratingsMaterials Forming and Machining: Research and Development Rating: 3 out of 5 stars3/5Composite Materials: Concurrent Engineering Approach Rating: 0 out of 5 stars0 ratingsBiofuels for a More Sustainable Future: Life Cycle Sustainability Assessment and Multi-Criteria Decision Making Rating: 0 out of 5 stars0 ratingsProduct Performance Evaluation using CAD/CAE: The Computer Aided Engineering Design Series Rating: 0 out of 5 stars0 ratingsDesign of Experiments for Engineers and Scientists Rating: 0 out of 5 stars0 ratingsProduct Manufacturing and Cost Estimating using CAD/CAE: The Computer Aided Engineering Design Series Rating: 4 out of 5 stars4/5Sustainable Engineering Products and Manufacturing Technologies Rating: 0 out of 5 stars0 ratingsProduct Development: A Structured Approach to Consumer Product Development, Design, and Manufacture Rating: 5 out of 5 stars5/5Mathematics Applied to Engineering Rating: 5 out of 5 stars5/5Mechanical Design Engineering Handbook Rating: 4 out of 5 stars4/5Chemical Engineering Process Simulation Rating: 4 out of 5 stars4/5Systems Engineering for Aerospace: A Practical Approach Rating: 0 out of 5 stars0 ratings4D Printing: Fundamentals and Applications Rating: 0 out of 5 stars0 ratingsSimulation and Optimization in Process Engineering: The Benefit of Mathematical Methods in Applications of the Chemical Industry Rating: 0 out of 5 stars0 ratingsChemical Engineering Design: Principles, Practice and Economics of Plant and Process Design Rating: 4 out of 5 stars4/5Mathematical Optimization Terminology: A Comprehensive Glossary of Terms Rating: 0 out of 5 stars0 ratingsAn Applied Guide to Process and Plant Design Rating: 4 out of 5 stars4/5Model Management and Analytics for Large Scale Systems Rating: 0 out of 5 stars0 ratingsMethods in Sustainability Science: Assessment, Prioritization, Improvement, Design and Optimization Rating: 0 out of 5 stars0 ratingsStochastic Global Optimization Methods and Applications to Chemical, Biochemical, Pharmaceutical and Environmental Processes Rating: 0 out of 5 stars0 ratingsSustainable Product Development: Tools, Methods and Examples Rating: 0 out of 5 stars0 ratingsStrategic Industry-University Partnerships: Success-Factors from Innovative Companies Rating: 4 out of 5 stars4/5Sustainable Manufacturing and Design Rating: 0 out of 5 stars0 ratingsUser Acceptance Testing: A step-by-step guide Rating: 3 out of 5 stars3/5Assessing and Measuring Environmental Impact and Sustainability Rating: 0 out of 5 stars0 ratingsIndustrial Process Scale-up: A Practical Innovation Guide from Idea to Commercial Implementation Rating: 0 out of 5 stars0 ratings
Mathematics For You
The Little Book of Mathematical Principles, Theories & Things Rating: 3 out of 5 stars3/5My Best Mathematical and Logic Puzzles Rating: 5 out of 5 stars5/5Algebra - The Very Basics Rating: 5 out of 5 stars5/5Quantum Physics for Beginners Rating: 4 out of 5 stars4/5Algebra I Workbook For Dummies Rating: 3 out of 5 stars3/5The Everything Guide to Algebra: A Step-by-Step Guide to the Basics of Algebra - in Plain English! Rating: 4 out of 5 stars4/5Basic Math & Pre-Algebra For Dummies Rating: 4 out of 5 stars4/5Mental Math Secrets - How To Be a Human Calculator Rating: 5 out of 5 stars5/5Real Estate by the Numbers: A Complete Reference Guide to Deal Analysis Rating: 0 out of 5 stars0 ratingsCalculus Made Easy Rating: 4 out of 5 stars4/5Flatland Rating: 4 out of 5 stars4/5Standard Deviations: Flawed Assumptions, Tortured Data, and Other Ways to Lie with Statistics Rating: 4 out of 5 stars4/5A Mind for Numbers | Summary Rating: 4 out of 5 stars4/5The Math of Life and Death: 7 Mathematical Principles That Shape Our Lives Rating: 4 out of 5 stars4/5Relativity: The special and the general theory Rating: 5 out of 5 stars5/5The Math Book: From Pythagoras to the 57th Dimension, 250 Milestones in the History of Mathematics Rating: 3 out of 5 stars3/5Game Theory: A Simple Introduction Rating: 4 out of 5 stars4/5Introducing Game Theory: A Graphic Guide Rating: 4 out of 5 stars4/5Geometry For Dummies Rating: 5 out of 5 stars5/5Summary of The Black Swan: by Nassim Nicholas Taleb | Includes Analysis Rating: 5 out of 5 stars5/5The Thirteen Books of the Elements, Vol. 1 Rating: 0 out of 5 stars0 ratingsIs God a Mathematician? Rating: 4 out of 5 stars4/5Mental Math: Tricks To Become A Human Calculator Rating: 5 out of 5 stars5/5Algebra I For Dummies Rating: 4 out of 5 stars4/5The Everything Everyday Math Book: From Tipping to Taxes, All the Real-World, Everyday Math Skills You Need Rating: 5 out of 5 stars5/5
Reviews for Computational Methods and Production Engineering
0 ratings0 reviews
Book preview
Computational Methods and Production Engineering - J. Paulo Davim
Computational Methods and Production Engineering
Research and Development
First Edition
J. Paulo Davim
University of Aveiro, Aveiro, Portugal
Table of Contents
Cover image
Title page
Copyright
List of contributors
About the editor
Preface
1: Parallel direct solver for finite element modeling of manufacturing processes
Abstract
Acknowledgments
1.1 Introduction
1.2 Brief review of standard gauss elimination
1.3 Structure of the parallel direct solver
1.4 Test cases and evaluation parameters
1.5 Results and discussion
1.6 Conclusions
Appendix
2: Optimal inspection/actuator placement for robust dimensional compensation in multistage manufacturing processes
Abstract
2.1 Introduction
2.2 State-space approach for modeling multistage assembly processes
2.3 Feed-forward predictive control
2.4 Optimal inspection/actuator placement for feed-forward control
2.5 Case study
2.6 Conclusions
Appendix
3: Numerical optimization strategies for springback compensation in sheet metal forming
Abstract
Acknowledgments
3.1 Introduction
3.2 Springback compensation strategies
3.3 Modeling strategies
3.4 Evaluation strategies
3.5 Optimization algorithms
3.6 Parameterization strategies
3.7 Case study: U-rail
3.8 Results and discussion
3.9 Conclusions
4: Finite element modeling of hot rolling: Steady- and unsteady-state analyses
Abstract
4.1 Introduction
4.2 Thermomechanical analysis of hot rolling: An overview
4.3 Work-roll and workpiece interface behavior
4.4 Constitutive equation for material model
4.5 Basic steps of FEM
4.6 Different approaches in FEM
4.7 Solution methods
4.8 Steady- and unsteady-state analyses of hot rolling
4.9 Concluding remarks
5: Numerical modeling methodologies for friction stir welding process
Abstract
5.1 Introduction to FSW
5.2 Modeling of FSW: Requirement and complexities
5.3 General steps for modeling a process
5.4 Modeling of FSW with Lagrangian analysis
5.5 Modeling of FSW with Eulerian analysis
5.6 Modeling of FSW with coupled Eulerian-Lagrangian (CEL) analysis
5.7 Comparison of modeling methods
5.8 Conclusion
6: Modeling of hard machining
Abstract
6.1 Introduction to hard machining
6.2 Numerical modeling of hard machining
6.3 Soft computing and statistical methods modeling of hard machining
6.4 Conclusions
7: Multiresponse optimization in wire electric discharge machining (WEDM) of HCHCr steel by integrating response surface methodology (RSM) with differential evolution (DE)
Abstract
7.1 Introduction
7.2 Experimental work
7.3 Response surface methodology
7.4 Results and discussion
7.5 Differential evolution (DE) optimization
7.6 Conclusions
Index
Copyright
Woodhead Publishing is an imprint of Elsevier
The Officers’ Mess Business Centre, Royston Road, Duxford, CB22 4QH, United Kingdom
50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States
The Boulevard, Langford Lane, Kidlington, OX5 1GB, United Kingdom
© 2017 Elsevier Ltd. All rights reserved.
No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Details on how to seek permission, further information about the Publisher's permissions policies and our arrangements with organizations such as the Copyright Clearance Center and the Copyright Licensing Agency, can be found at our website: www.elsevier.com/permissions.
This book and the individual contributions contained in it are protected under copyright by the Publisher (other than as may be noted herein).
Notices
Knowledge and best practice in this field are constantly changing. As new research and experience broaden our understanding, changes in research methods, professional practices, or medical treatment may become necessary.
Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using any information, methods, compounds, or experiments described herein. In using such information or methods they should be mindful of their own safety and the safety of others, including parties for whom they have a professional responsibility.
To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any liability for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions, or ideas contained in the material herein.
Library of Congress Cataloging-in-Publication Data
A catalog record for this book is available from the Library of Congress
British Library Cataloguing-in-Publication Data
A catalogue record for this book is available from the British Library
ISBN: 978-0-85709-481-0 (print)
ISBN: 978-0-85709-482-7 (online)
For information on all Woodhead publications visit our website at https://www.elsevier.com/books-and-journals
Publisher: Glyn Jones
Acquisition Editor: Glyn Jones
Editorial Project Manager: Lindsay Lawrence
Production Project Manager: Stalin Viswanathan
Cover Designer: Matthew Limbert
Typeset by SPi Global, India
List of contributors
J.V. Abellán-Nebot Universitat Jaume I, Castelló de la Plana, Spain
A. Andrade-Campos University of Aveiro, Aveiro, Portugal
E. Ferreira
University of Aveiro, Aveiro
University of Coimbra, Coimbra, Portugal
V.N. Gaitonde B.V.B. College of Engineering and Technology, Hubli, India
Rahul Jain Indian Institute of Technology, Kharagpur, India
N.E. Karkalos National Technical University of Athens, Athens, Greece
S.R. Karnik B.V.B. College of Engineering and Technology, Hubli, India
J. Liu The University of Arizona, Tucson, AZ, United States
A. Maia University of Aveiro, Aveiro, Portugal
M. Manjaiah GeM Lab, Nantes, France
S. Maradi B.V.B. College of Engineering and Technology, Hubli, India
A.P. Markopoulos National Technical University of Athens, Athens, Greece
P.A.F. Martins Universidade de Lisboa, Lisbon, Portugal
L.F. Menezes University of Coimbra, Coimbra, Portugal
C.V. Nielsen Technical University of Denmark, Lyngby, Denmark
M.C. Oliveira University of Coimbra, Coimbra, Portugal
Surjya K. Pal Indian Institute of Technology, Kharagpur, India
J. Paulo Davim University of Aveiro, Aveiro, Portugal
I. Peñarrocha Universitat Jaume I, Castelló de la Plana, Spain
P.M. Petkar B.V.B. College of Engineering and Technology, Hubli, India
Matruprasad Rout Indian Institute of Technology, Kharagpur, India
E. Sales-Setién Universitat Jaume I, Castelló de la Plana, Spain
Shiv B. Singh Indian Institute of Technology, Kharagpur, India
About the editor
J. Paulo Davim received his PhD in Mechanical Engineering in 1997, MSc degree in Mechanical Engineering (materials and manufacturing processes) in 1991, Dipl.-Ing Engineer’s degree (5 years) in Mechanical Engineering in 1986, from the University of Porto (FEUP), the Aggregate title (Full Habilitation) from the University of Coimbra in 2005 and a DSc from London Metropolitan University in 2013. He is Eur Ing by FEANI-Brussels and Senior Chartered Engineer by the Portuguese Institution of Engineers with a MBA and Specialist title in Engineering and Industrial Management. Currently, he is Professor at the Department of Mechanical Engineering of the University of Aveiro, Portugal. He has more than 30 years of teaching and research experience in Manufacturing, Materials, and Mechanical Engineering with special emphasis in Machining and Tribology. He has also interest in Management/Industrial Engineering and Higher Education for Sustainability/Engineering Education. He has received several scientific awards. He has worked as evaluator of projects for international research agencies as well as examiner of PhD thesis for many universities. He is the Editor in Chief of several international journals, Guest Editor of journals, books Editor, book Series Editor, and Scientific Advisory for many international journals and conferences. Presently, he is an Editorial Board member of 30 international journals and acts as reviewer for more than 80 prestigious Web of Science journals. In addition, he has also published as editor (and coeditor) of more than 100 books and as author (and coauthor) of more than 10 books, 80 book chapters, and 400 articles in journals and conferences (more than 200 articles in journals indexed in Web of Science/h-index 37+ and SCOPUS/h-index 45+).
J. Paulo Davim, University of Aveiro, Aveiro, Portugal
Preface
Production engineering is a branch of engineering that involves the design, development, implementation, operation, maintenance, and control of all processes in the manufacture of a product.
It is an interdisciplinary subject requiring the collaboration of individuals trained in manufacturing engineering, industrial engineering, product design, management, etc.
Recently, there has been increased interest in developing computational methods to be applied in production engineering. Consequently, in production engineering, computational methods have achieved several applications, namely, modeling manufacturing processes, monitoring and control, parameters optimization and computer-aided process planning, etc.
This research book aims to provide information on computational methods and production engineering for modern industry. The initial chapter of the book provides parallel direct solver for finite element modeling of manufacturing processes. Chapter 2 is dedicated to optimal inspection/actuator placement for robust dimensional compensation in multistage manufacturing processes. Chapter 3 presents numerical optimization strategies for springback compensation in sheet metal forming. Chapter 4 covers finite element modeling of hot rolling (steady- and unsteady-state analyses). Chapter 5 is dedicated to numerical modeling methodologies for friction stir welding process. Chapter 6 contains information on modeling of hard machining. Finally, the last chapter of the book is dedicated to multiresponse optimization in wire electric discharge machining (WEDM) of HCHCR steel by integrating response surface methodology (RSM) with differential evolution (DE).
The present book can be used as a research book for final undergraduate engineering course or as a topic on computational methods and production engineering at the postgraduate level. This book can serve as reference for academics, researchers, manufacturing, mechanical and industrial engineers, as well as professionals in production engineering. Also, this book presents scientific interest for industry, centers of the research, laboratories, and universities throughout the world.
The editor acknowledges Woodhead/Elsevier for this opportunity and for their professional support. Finally, I would like to thank all the chapter authors for their availability for this work.
J. Paulo Davim, University of Aveiro, Aveiro, Portugal
1
Parallel direct solver for finite element modeling of manufacturing processes
C.V. Nielsen*; P.A.F. Martins† * Technical University of Denmark, Lyngby, Denmark
† Universidade de Lisboa, Lisbon, Portugal
Abstract
This chapter describes and evaluates the parallelization of a direct equation solver for implementation in the finite element modeling of manufacturing processes. The parallel solver is implemented in an existing three-dimensional finite element computer program for problems including large deformations, contact development, and electrical and thermal fields. Speed-up and efficiency are evaluated on a standard personal computer. Considerable time savings are achieved, allowing computations to take place in industry within reasonable computation time without access to any kind of supercomputers or clusters. The direct equation solver is parallelized for shared memory platforms by OpenMP instructions, and it can be directly implemented in existing finite element codes utilizing skyline matrix storage, no matter the physical dimensions of the problems dealt with. Only the call to the existing solver has to be substituted by a call to the proposed parallel solver. The modification of the parallel direct equation solver to be applied as preconditioner of iterative equation solvers is also discussed. The FORTRAN source code is provided in the Appendix.
Keywords
Direct solver; Parallelization; OpenMP; FEM; Numerical methods
Acknowledgments
The authors would like to acknowledge the support provided by Fundação para a Ciência e a Tecnologia of Portugal and IDMEC under LAETA—UID/EMS/50022/2013 and PDTC/EMS-TEC/0626/2014.
1.1 Introduction
The central processing unit (CPU) time is of paramount importance in finite element modeling of manufacturing processes. Because the most significant part of the CPU time is consumed in solving the main system of equations resulting from finite element assemblies, different approaches have been developed to optimize solutions and reduce the overall computational costs of large finite element models.
The simplest approach is to apply faster solution techniques by replacing direct equation solvers by iterative equation solvers. There are various types of iterative solvers but the conjugate gradient (CG) iterative solvers proposed by Lanczos (1952) and Hestenes and Stiefel (1952) are among the simplest and more widely used in finite element computer programs.
Despite the advantages of CG iterative solvers, there are two major concerns (and challenges) related to its utilization in finite element modeling of manufacturing processes. First, precision is lost compared to that of direct solvers, because final accuracy depends on the threshold value utilized for accepting the solution, which inevitably results from a compromise between the desired accuracy and the required CPU time. In case small inaccuracies accumulate during finite element modeling of a manufacturing process involving a large number of solution steps, they may lead to poor satisfaction of the boundary conditions and to inaccuracies in fulfilling symmetry conditions. For instance, a zero displacement associated with a symmetry line (or plane) may be computed as a very small nonzero displacement creating problems in the overall final accuracy of the numerical simulation. In problems involving contact the earlier mentioned problems may, for larger threshold values, also disturb the contact algorithms, eventually leading to penetration in contact pairs.
The second drawback results from the fact that CG iterative solvers suffer from low robustness and applicability in case of ill-conditioned equation systems. In fact, iterative solvers have been reported unstable when dealing with ill-conditioned equation systems, whereas direct solvers have proved to be more robust (Farhat and Wilson, 1988). This is particularly relevant for finite element formulations that make use of penalties (i.e., very large numbers) to impose material incompressibility and/or to handle contact between different objects (e.g., deformable vs. deformable objects and/or deformable vs. rigid objects) and for simulations with large rigid body motion (e.g., when a preform is settling into a metal forming die undergoing very little deformation).
Among various improvements to optimize convergence and to improve the CPU time of CG iterative solvers, preconditioning and parallelization have been the most widely used techniques (Meijerink and van der Vorst, 1977).
A more complex approach to optimize solutions and reduce the overall computational cost of large finite element models has been achieved by decomposition of a finite element model into subdomains that are simultaneously solved by means of parallel computation (Kim and Im, 2003). The interface nodes between subdomains require communication between different processors and, therefore its overall number should be minimized in order to avoid computational bottlenecks. Farhat (1988) and Al-Nasra and Nguyen (1992) were among the first researchers to explore the trade-off between the number of interface nodes and subdomain size and to propose algorithms for optimal decompositions.
Parallelization of direct solvers is usually not considered as an alternative to parallelization of CG iterative solvers because it requires large amount of data communication between processors and also because it is considered more tedious and difficult to implement in existing finite element computer programs. However, parallelization of direct solvers may be an appropriate solution for two-dimensional finite element analysis of manufacturing processes, for three-dimensional analysis of manufacturing processes involving medium size models, and for preconditioning CG iterative solvers. In addition to what is mentioned earlier, it is worth noting that the solution accuracy of parallel direct solvers is identical to that of sequential direct solvers and that parallel direct solvers can also be utilized with solution decomposition approaches regardless of the problem to be solved.
Farhat and Wilson (1988) were among the first researchers to present parallel direct solvers. Parallelization can be applied for local memory processors as well as for shared memory processors, where the first is typically applied to a cluster of multiple computers, whereas the latter typically would be one computer with multiple threads.
Synn and Fulton (1995) were also among the first researchers to propose evaluation procedures to predict the performance of parallel direct solvers using skyline matrix storage. The usage of skyline or more efficient compressed sparse row storage formats is very often utilized in computer programs as it explores the large sparsity of the equation systems that are typical of finite element models in manufacturing.
In the present chapter a parallel direct solver for shared memory processors is presented and evaluated. The implementation is carried out in a personal computer (PC) due to increasing requirements from industry for running finite element simulations in standard hardware equipped with several cores and threads with shared memory. It is therefore an obvious request that the finite element computer programs can utilize multiple threads to reduce the computation time.
The direct solver to be presented is parallelized by OpenMP instructions in a FORTRAN implementation. The structure of the solver is explained in the chapter, and its performance is evaluated in terms of speed-up and efficiency. Amdahl's law (Hill and Marty, 2008) is applied to evaluate the amount of the code being parallel and to evaluate the potential benefit associated with an increasing number of threads. The entire source code is provided in the Appendix for the benefit of those who develop finite element computer programs. Only the call to the solver in existing computer programs using the same matrix storage format has to be replaced by a call to the presented solver. The requested inputs are the stiffness matrix in appropriate storage format together with the corresponding pointers to the diagonal positions, the right-hand side, the number of equations, and the number of threads to be utilized. A special implementation of the parallel direct solver that can be effectively and efficiently utilized as a preconditioner of parallel CG iterative solvers is also discussed.
A benchmark test case and a resistance spot welding test case selected from an industrial application enrich the overall presentation. The industrial example requires solution of a coupled electro-thermo-mechanical model in three dimensions with contact between individual hexahedral meshes. The case is further complicated by having large temperature gradients across the contacting interfaces, and therefore also large gradients in the mechanical stress response.
1.2 Brief review of standard gauss elimination
1.2.1 Skyline matrix storage
The presentation of the parallel direct solver is focused on the equation systems resulting from typical finite element assemblies,
(1.1)
where [Amatrix and {x} and {bvectors containing the unknowns and the right-hand side, respectively.
Due to symmetry of the system matrix, only half of the matrix needs to be built and stored (slightly more than half due to storage of all diagonal positions). Furthermore, since most finite element equation systems are sparse and by proper node numbering have many zeros far from the diagonal, it is common to employ skyline storage to reduce memory requirements by omitting all zeros above the highest skyline positions (Fig. 1.1). Zeros may still exist below this envelope as the skyline encloses all nonzero positions.
Fig. 1.1 System matrix storage by utilizing symmetry (left) and skyline format by omitting zeros (right). Only the grayed positions are stored. The dashed line covering the highest skyline positions indicates the requirements in case of banded matrix storage format.
In skyline matrix storage, the equation system is typically stored in a one-dimensional vector {s} with an additional index vector {i} pointing to the diagonal positions. This is illustrated in Fig. 1.2 up to the 7th column. The size of the skyline vector will be the number of positions under the skyline. The size of the index vector equals the number of rows or columns n. Then, it follows that the size of the skyline vector is in as the last diagonal is the last position in the skyline vector.
Fig. 1.2 Format of skyline vector { s } and index vector { i } based on the original system matrix [ A ]. Numbers correspond to the position in the skyline vector.
1.2.2 Gauss elimination
Among various solution techniques used by direct solvers, Gauss elimination was chosen due to its wide utilization and adequacy for parallelization. The solution of an equation system by Gauss elimination with column reduction comprises the following three basic steps:
• factorization of the system matrix and reduction of the right-hand side vector (this step is performed column by column, thereby being with column reduction
)
• division of the right-hand side vector by the diagonals of the system matrix
• backward substitution.
1.2.2.1 Factorization of the system matrix and reduction of the right-hand side vector
The first step in the factorization of the system matrix is illustrated by Fig. 1.3A for column j have been processed. The number of operations equals the active column height −2 as the diagonal position and the topmost position are not processed. The illustration in Fig. 1.3A has active column height 7 and five operations are depicted in the subfigures. In each operation, the kth position is subtracted from the dot product formed by the vectors marked by o
s and x
s, i.e.
Fig. 1.3 Factorization of system matrix and reduction of right-hand side. (A) Reduction of off-diagonal positions in column j by subtraction of dot product formed by o
s and x
s. (B) Reduction of diagonal positions in system matrix and reduction of right-hand side vector.
(1.2)
Hereafter, all positions above the diagonal are divided by the diagonal position in the same row. In Fig. 1.3B, this corresponds to
(1.3)
where division is position-wise. This is followed by reduction of the diagonal term by
(1.4)
which means subtraction of each multiplication of new and old off-diagonal position. New is defined as "after Eq. (1.3) and old is defined as
before Eq. (1.3)."
Finally, the jth position in the right-hand side vector is reduced by subtraction of the dot product spanned by the marked x
s and v
s in Fig. 1.3B. Note that the positions marked by x
s are now the latest updated, meaning after (1.4). The reduction of the right-hand side vector can be written as
(1.5)
1.2.2.2 Division of right-hand side vector by system matrix diagonals
This step is straight forward: Each position, j, in the right-hand side vector is divided by the system matrix diagonal term from the jare eliminated from the jth row.
1.2.2.3 Backward substitution
The unknowns are now found by backward substitution and stored in the right-hand side vector {b}. The right-hand side vector is processed backward, such that Fig. 1.4 illustrates the substitution for the jalready processed. The positions marked by v
s in the figure are modified by subtraction of positions in the system matrix marked by x
s in the jth column. Before subtraction, these positions are multiplied by bj, the jth position in the right-hand side vector, that is
Fig. 1.4 in right-hand side vector by subtraction of system matrix positions marked by x
s multiplied by the j th position of the right-hand side vector.
(1.6)
.
1.3 Structure of the parallel direct solver
The parallelization of the direct solver with skyline matrix storage is explained in this section with variable names matching the source code included in the Appendix. The factorization of the system matrix and reduction of the right-hand side (Section 1.2.2.1) is parallelized, whereas the division of the right-hand side vector by the system matrix diagonals (Section 1.2.2.2) as well as the backward substitution (Section 1.2.2.3) is left sequential as the time spent on these tasks is marginal compared to the factorization and reduction. The remaining of this section is devoted to the description of the parallelized part of the solver.
A few variables are introduced preliminary:
1.3.1 Parallel region
The factorization and reduction are performed within a parallel region defined by